We are now the AI Security Institute

Please enable javascript for this website.

A

A

Research

Improving our understanding of advanced AI

Our mission is to equip governments with a scientific understanding of the risks posed by advanced AI.

View research agenda

Conversational AI increases political knowledge as effectively as self-directed internet search

Human Influence

•

Sep 8, 2025

Lessons from studying two-hop latent reasoning

Science of Evaluations

•

Sep 6, 2025

Deep ignorance: Filtering pretraining data builds tamper-resistant safeguards into open-weight LLMs

Safeguards

•

Aug 8, 2025

Security challenges in AI agent deployment: Insights from a large scale public competition

Safeguards

•

Jul 27, 2025

STACK: Adversarial attacks on LLM safeguard pipelines

Safeguards

•

Jul 18, 2025

The levers of political persuasion with conversational AI

Human Influence

•

Jul 18, 2025

Chain of thought monitorability: A new and fragile opportunity for AI safety

Control

•

Jul 15, 2025

HiBayES: A hierarchical bayesian modelling framework for AI evaluation statistics

Science of Evaluations

•

Jul 13, 2025

White Box Control at UK AISI - update on sandbagging investigations

Control

•

Jul 10, 2025

Lessons from a chimp: AI "scheming" and the quest for ape language

Science of Evaluations

•

Jul 4, 2025

An example safety case for safeguards against misuse

Safety Cases

•

Jun 5, 2025

Existing Large Language Model unlearning evaluations are inconclusive

Science of Evaluations

•

May 31, 2025

RepliBench: Evaluating the autonomous replication capabilities of language model agents

Science of Evaluations

•

May 21, 2025

How to evaluate control measures for LLM agents? A trajectory from today to superintelligence

Control

•

May 7, 2025

An alignment safety case sketch based on debate

Safety Cases

•

May 6, 2025

Evaluating explanations: An explanatory virtues framework for mechanistic interpretability

Control

•

May 2, 2025

A mathematical philosophy of explanations in mechanistic interpretability

Control

•

May 1, 2025

AgentHarm: A benchmark for measuring harmfulness of LLM agents

Science of Evaluations

•

Apr 18, 2025

Model tampering attacks enable more rigorous evaluations of LLM capabilities

Safeguards

•

Mar 2, 2025

Fundamental limitations in defending LLM finetuning APIs

Safeguards

•

Feb 20, 2025

Safety Cases: A scalable approach to Frontier AI safety

Safety Cases

•

Feb 10, 2025

Emerging practices in frontier AI safety frameworks

Governance

•

Feb 5, 2025

Why human-AI relationships need socioaffective alignment

Societal Resilience

•

Feb 4, 2025

Principles for evaluating misuse safeguards of frontier AI systems

Safeguards

•

Feb 4, 2025

A sketch of an AI control safety case

Safety Cases

•

Jan 28, 2025

Safety case template for frontier AI: A cyber inability argument

Safety Cases

•

Dec 12, 2024